Automatic Salient Regions of Interest Extraction Based on Edge and Region Integration

A major problem in content-based image retrieval is the unsupervised identification of perceptually salient regions of interest in images. This paper proposed a scheme for automatic extracting the salient regions of interest in images. In this scheme, first classify images to two different groups according to their wavelet modulus maxima point densities, and then extract the salient regions of interest differently. The approach of extraction combines the wavelet modulus maxima edges with the color regions information of mean shift based segmentation. Experiments testify that the proposed scheme is effective especially for low depth of field images and landscape images